Launch Linkedin Campaign

Launch LinkedIn Campaign

I'll wrap two CLI commands — campaign:create (creates the campaign + pulls qualified leads from the holding pool) and campaign:create-sequence (drafts the connect→DM1→DM2 sequence). Both side-effecting → both shell-out per the 0.13.0 architecture.

When This Skill Applies

Use this skill when the user says:

• "launch a LinkedIn campaign for these leads"
• "send a LinkedIn outreach to this list"
• "start the outbound to the qualified leads"
• "run the LinkedIn sequence on this result set"
• "fire the connect-then-DM flow"

NOT this skill (use qualify-leads instead):

• "score these leads first" / "qualify the engagers" — qualification runs the 7-gate pipeline. Run that BEFORE this skill.

NOT this skill (use personalize-message instead):

• "personalize a single DM" — that's per-lead copy. This skill handles bulk sequence generation.

NOT this skill (use scrape-post-engagers instead):

• "pull who liked this post" — that produces a result set. This skill consumes a holding pool of qualified leads.

What This Skill Does

1. Hypothesis gate. Before doing anything, checks that the outbound hypothesis is recorded — ~/.gtm-os/frameworks/installed/outreach-campaign-builder.hypothesis.json must exist. If missing: refuses to launch and routes the user to framework:set-hypothesis (or setup Step 10).
2. Asks for campaign title + optional leads filter + sequence YAML + source CSV/JSON path.
3. Shells out to campaign:create to write the campaign row + pull qualified leads from the holding pool.
4. Shells out to campaign:create-sequence to draft the connect→DM1→DM2 sequence.
5. Renders both results + asks "ready to send?" — does not auto-trigger sending.
6. Suggests enabling campaign:track cron (or running it manually) for monitoring.

Pre-flight (before Step 1)

Onboarding interruption guard

test -f ~/.gtm-os/.in-flight-setup && echo "BLOCKED" || echo "OK"

If BLOCKED, stop. Tell the user to finish yalc-gtm start first.

Hypothesis gate (THE critical check)

test -f ~/.gtm-os/frameworks/installed/outreach-campaign-builder.hypothesis.json && echo "OK" || echo "MISSING"

If MISSING, refuse to proceed:

"Can't launch a LinkedIn campaign without a recorded outbound hypothesis. Either:

(a) finish setup Step 10 to record one (yalc-gtm start --review-in-chat and re-run setup), or

(b) record one directly:

yalc-gtm framework:set-hypothesis outreach-campaign-builder \
  --icp-segment '<segment>' \
  --message-angle '<angle>' \
  --signal-trigger '<signal>' \
  --expected-reply-rate 0.05

Then re-invoke me."

This guard is enforced by the skill, not the CLI (campaign:create does NOT check the sidecar today).

Workflow

Step 0 — Ask for inputs

One question at a time:

1. Campaign title? (e.g., "VP Marketing Q2 outbound — segment-fit hire signal")
2. Lead source? Two paths:
   • Use the holding pool of all qualified leads (default — pass no --leads-filter).
   • Filter to a subset via --leads-filter '<json-shape>' (e.g., '{"score":{"$gte":80}}').
3. Sequence YAML path? (--sequence <path> — required by campaign:create-sequence). Usually configs/sequences/connect-dm1-dm2.yaml or a custom path.
4. Source CSV/JSON? (--source <path> — required by campaign:create-sequence). The leads file the sequence will personalize against.
5. Optional toggles:
   • --linkedin-account <id> (multi-account routing)
   • --timezone <tz> (default Pacific or whatever the framework default is)
   • --start-at <date> (default = now)
   • --send-window '<HH:MM-HH:MM>'
   • --active-days <mon,tue,wed,thu,fri>
   • --delay-mode <natural|fast>
   • --dry-run (preview without writing to Unipile)

Step 1 — Read the hypothesis sidecar for the --hypothesis arg

HYPOTHESIS=$(jq -r '.message_angle' ~/.gtm-os/frameworks/installed/outreach-campaign-builder.hypothesis.json)

The CLI's --hypothesis flag accepts a string. Pass the message_angle so campaign-intelligence can score the actual reply rate against the declared expected rate later.

Step 2 — Shell out to campaign:create

cd ~/Desktop/gtm-os && set -a && source .env.local && set +a && \
  npx tsx src/cli/index.ts campaign:create \
    --title "<title>" \
    --hypothesis "$HYPOTHESIS"

(Add --leads-filter '<json>', --auto-copy, --segment-id, --timezone, --start-at, --send-window, --active-days, --delay-mode, or --dry-run if the user opted in.)

Per the 0.13.0 benchmark: side-effecting commands always shell out. The CLI's withDiagnostics() wrapper handles env loading + tenant resolution.